Sample report
A real report, with the company's name taken out
This is one of the reports we send after the free check. It is real: we ran it on 2 September 2026 for an HVAC company in the Atlanta area. Every number below is exactly as we measured it. What has been removed is who it was about.
Why nobody is named
We do not publish a company's name next to a finding that AI assistants are not recommending it. Not with permission, not with the company's blessing — not at all. So the company, its city, and every competitor the assistant named have been replaced. The city is given as "the Atlanta area" rather than the suburb it is in, because a single suburb plus a single trade is a short enough list to start guessing at.
Nothing else was changed. The counts, the dates, the questions and the wording of the findings are the report as it was generated.
What the report found
We asked ChatGPT 34 questions a buyer in their area would realistically ask, 5 times each, and recorded which companies it named. That is 170 answers.
0 of 130
buyer questions where the assistant was not given their name, and named them anyway
40 of 40
questions that contained their name, where the assistant had something to say about them
Across every question where the buyer did not already know their name, this company was never mentioned. Handed the name directly, the assistant had something to say every single time. This is a question of being found, not of being unknown once someone looks.
Where they appear, and where they don't
| Question type | Named | Asked | Rate |
|---|---|---|---|
| Urgent needSomeone with a problem right now, asking who to call. | 0 | 40 | 0% |
| ComparisonSomeone weighing their options and asking who is best. | 0 | 50 | 0% |
| Brand checkSomeone who already has the name and is checking the company out. | 40 | 40 | 100% |
| Price researchSomeone asking what this kind of work should cost. | 0 | 40 | 0% |
Brand check is the row to read differently. Those questions contain the company's name, so a mention there shows the assistant has something to say once it is asked — not that a new customer would ever find them. The other three rows are the ones a stranger asks, and those are the ones that came back at zero.
What it said when it was handed the name
These come from the forty answers to questions that contained the company's name. We have edited them. Out came the city, the county, the state's contractor-license names and the second trade this company sells — details specific enough to identify them. What the assistant actually did is unchanged, and in more than half of the forty it did the same thing: it refused to vouch for the company at all, saying it could not confirm their reputation, their license status or their reviews without checking live sources. The rest describe what the company sells, or set conditions on recommending it — like the last one below.
- Could not confirm their current reputation, licensing status, or recent review pattern without live verification. Suggested checking their state contractor license, insurance, recent reviews, a written estimate, and permits before hiring.
- Mentioned as the business the user asked about. Said it could not summarize current reviews without live access.
- Declined to judge whether their reviews were genuine without examining the review profiles, the dates, and the reviewer histories.
- Could not verify whether they are consistently more expensive than other companies in the area without comparing current written estimates for the same job.
- Worth asking for a quote and putting on a shortlist — but only hire them if their proposal is thorough and competitively priced against two other licensed local installers, and if they pull local permits and offer a meaningful labor warranty.
One more finding did not survive the edit but is worth stating plainly, because it is the sort of thing this report exists to catch. Asked the same question about how long the company had been in business, five answers gave three different founding years. The exact years are left out here for the same reason as everything else on this page.
Who gets named instead
This is the part an owner forwards to their team. The assistant was not short of companies to recommend in this area — it named 71 of them across the run. These are the eight it named most often.
| Company | Times named | Share of answers |
|---|---|---|
| Company A | 68 | 40% |
| Company B | 67 | 39% |
| Company C | 64 | 38% |
| Company D | 47 | 28% |
| Company E | 43 | 25% |
| Company F | 39 | 23% |
| Company G | 34 | 20% |
| Company H | 25 | 15% |
Plus 63 more named less often. In your report these are real company names, and they are the reason the report is worth reading. Here they are letters, for the same reason the subject is.
One honest caveat about this table: a company sometimes appears twice, because an assistant writes the same business name differently in different answers and the count follows the name as written. We report it that way rather than quietly merging rows and changing the numbers.
How this was measured
- Questions asked
- 17034 questions × 5 repetitions each
- Assistant
- ChatGPTmodel gpt-5.6-terra
- Date collected
- Question set
- v4so a later report can be compared to this one
- Requests that failed
- 0a failed request is excluded, never counted as "named nobody"
- Subject
- An HVAC companyin the Atlanta area
This is a frequency count, not a ranking. There is no score here and no position out of a hundred, because neither would be a real thing. It is how often something happened, out of how many times we asked.
AI assistants are non-deterministic — the same question asked twice can return different companies — and answers are partly personalized to the person asking. That is why every question here was asked five times and the result is reported as a frequency rather than as one answer treated as the truth.
This measures what the assistant says on its own. Advertising does not change it. Ads in these assistants are labeled and sit separately from the answer itself. Getting named in the answer is different work: reviews, your Google Business Profile, consistent business information across directories, and being present in the sources these models draw on. Most of that is work you do, not work an agency does for you.
We can show you that a number like this changed. We cannot prove what caused it to change — the models move for reasons unrelated to anyone's work. Anyone telling you otherwise is selling something. This report is a measurement taken once. It is not a before-and-after, and no work of ours had been done on this company.
One limitation we are still checking. We ask the assistant through its interface for developers, which answers from what the model already knows rather than searching the web while it replies. Consumer ChatGPT often does search the web on local questions, so what a buyer sees may differ from what we measured. You can see the model saying so itself in the answers above, where it repeatedly declined to confirm anything "without live access". We are working out how much difference that makes, and we will publish what we find either way.
One company is not a study. This page is an example of what a report looks like, not evidence about an industry or about anyone else's numbers. Yours will be different, and the only way to know how is to run it.
Your turn
Now do it on your site
Start with the instant check — five checks run live against your site, on the page, no email needed. The report above is what comes back after that, along with the six deeper checks on your website, within one business day.
Run the free instant check